This server brings token measurement and conflict detection to your AI context files. It exposes tools to scan directories for .cursorrules, CLAUDE.md, AGENTS.md, and other context formats, then report token counts, flag contradictions between rules, and highlight duplicates that waste budget. You get lint scoring with CI-friendly exit codes, task-based rule selection that respects token limits, and sync operations to maintain a single source of truth across multiple IDE formats. Reach for this when you're juggling context files across Cursor, Claude, and Copilot and need to know what's actually being injected into each conversation, or when your agent performance degrades and you suspect bloated or conflicting instructions are the culprit.
claude mcp add --transport stdio ofershap-ai-context-kit uvx ai-context-kit